I've watched Product teams make the same mistake more times than I can count. Someone gets excited about an idea, the enthusiasm spreads, and suddenly there's an all-hands push to build a new feature.

Other projects get quietly deprioritized. People stay late. Resources get stretched. And then, somewhere after everything ships and the initial excitement dies down, the team finally pauses long enough to look at the data... and realizes the thing they just built isn't something users actually wanted.

The question almost nobody asks — at least not until it's too late — is: should we build this at all?

I was first introduced to a framework for answering that question by JW Kim, who was my manager at Givelify and one of the sharpest product thinkers I've had the privilege of working with. JW pointed me toward a book called The Right It by Alberto Savoia, who was Google's first Engineering Director. The core argument stuck with me immediately: at least 80% of new products fail in the market. Not because they were built poorly, but because they were the wrong thing to build in the first place. Savoia calls it the Law of Market Failure, and it applies to startups and Fortune 500s equally.

What's changed — significantly, and recently — is how cheap and fast it's become to actually do something about it.

The Framework: Pretotyping

Savoia's solution is something he calls pretotyping (not a typo — it comes before prototyping, and the distinction matters).

A prototype tests whether something can be built and how it should work. A pretotype tests something far more fundamental: whether anyone actually wants it. The goal is to answer that question as quickly and cheaply as possible, before a single sprint of real development begins.

The most famous example is one I've repeated in a lot of conversations: Jeff Hawkins, the founder of Palm, carved a wooden block roughly the size of the device he was imagining, drew an interface on a slip of paper, and carried it around for weeks before committing to build anything. He'd pull it out in meetings. Pretend to take notes. Check his "calendar." He wasn't testing functionality — he was testing his own behavior. Would he actually reach for this thing? Would it fit naturally into how he worked?

It did. The Palm Pilot became one of the defining personal devices of its era.

IBM did something similar — and even more theatrical — to test early interest in speech-to-text software before the technology was actually feasible. A human typist, hidden in an adjacent room, transcribed speech in real time while users believed they were interacting with a working AI system. The reactions those users had told IBM everything they needed to know about whether the concept had commercial legs — without building a single line of production code.

Fake it before you make it. Build the smallest possible simulation of the experience, put it in front of real people, and watch what they actually do.

Skin in the Game

Here's the piece of Savoia's framework I find myself thinking about most, because it cuts directly against one of the things teams rely most heavily on: survey data.

Opinions are free. If you ask someone "would you use a feature that does X?" they will almost always say yes, because agreeing costs them nothing. Savoia calls this the "no skin in the game" problem — and it's why focus group results, surveys, and stakeholder enthusiasm are all notoriously unreliable predictors of what people will actually do.

People tell you what they think you want to hear, or what they think they'd want, or just whatever answer gets them out of the conversation fastest.

Useful pretotyping data requires the opposite: someone has to give something up to signal real interest. Time spent engaging with a prototype they could've ignored. An email address. A click-through on a "Sign up" button that doesn't go anywhere yet. Money, in the strongest version of the approach.

The Tesla Model 3 reservation program is the clearest modern example of this I know. In 2016, Tesla didn't survey people about whether they'd consider buying an electric car someday. They asked people to put down a $1,000 deposit on a car that didn't exist yet, that most of them had never seen in person, with delivery estimated more than a year out. Over 325,000 people did it in the first week. That number carried real weight precisely because it required a considered decision rather than a reflexive answer to a question. (Worth noting: the $1,000 deposit was fully refundable, and roughly 23% of those reservations were eventually cancelled. But it's still a far more reliable signal than any survey would've produced.)

A poll asking "do you want an electric car someday?" would have produced a bigger number and told you almost nothing useful.

Savoia calls the good stuff YODA — Your Own Data: collected firsthand, recently, with something at stake. The higher the stakes attached to collecting it, the more you can trust it. And critically, it doesn't have to be money. Getting someone to spend 20 minutes with a prototype they could've declined to engage with? That's real data too.